[Paper Review] Impact of Data Normalization on Deep Neural Network for Time Series Forecasting

Samit Bhanja, Abhishek Das|arXiv (Cornell University)|Dec 13, 2018

Stock Market Forecasting Methods14 references102 citations

TL;DR

The paper investigates how various data normalization techniques affect Deep Recurrent Neural Network forecasting of time series, applying DRNN to BSE and NYSE closing indices.

ABSTRACT

For the last few years it has been observed that the Deep Neural Networks (DNNs) has achieved an excellent success in image classification, speech recognition. But DNNs are suffer great deal of challenges for time series forecasting because most of the time series data are nonlinear in nature and highly dynamic in behaviour. The time series forecasting has a great impact on our socio-economic environment. Hence, to deal with these challenges its need to be redefined the DNN model and keeping this in mind, data pre-processing, network architecture and network parameters are need to be consider before feeding the data into DNN models. Data normalization is the basic data pre-processing technique form which learning is to be done. The effectiveness of time series forecasting is heavily depend on the data normalization technique. In this paper, different normalization methods are used on time series data before feeding the data into the DNN model and we try to find out the impact of each normalization technique on DNN to forecast the time series. Here the Deep Recurrent Neural Network (DRNN) is used to predict the closing index of Bombay Stock Exchange (BSE) and New York Stock Exchange (NYSE) by using BSE and NYSE time series data.

Motivation & Objective

• Motivate the need to redefine DNN models for nonlinear, dynamic time series forecasting.
• Evaluate how different data normalization techniques impact learning and forecasting performance.
• Assess pre-processing, network architecture, and parameter choices before feeding data into DRNN models.

Proposed method

• Apply different normalization methods to time series data before DRNN input.
• Use Deep Recurrent Neural Network to predict the closing index of BSE and NYSE.
• Compare the impact of normalization methods on DRNN forecasting outcomes.

Experimental results

Research questions

• RQ1 How do different data normalization techniques affect DRNN-based time series forecasting?
• RQ2 Which normalization method provides the most effective forecasting performance for BSE and NYSE data?
• RQ3 What is the role of data normalization in shaping the learning dynamics of DRNN for nonlinear time series?

Key findings

• The paper explores the impact of normalization techniques on DRNN forecasting performance for time series data.
• No specific numerical results are provided in the abstract excerpt.